| Oracle FAQ | Your Portal to the Oracle Knowledge Grid | |
 |
 | |||
| Oracle FAQ | Your Portal to the Oracle Knowledge Grid | |
|
| |||
Home -> Community -> Mailing Lists -> Oracle-L -> Re: I/O buffer cache
Hi, novice (I'm reluctant to call you that),
On UNIX and Linux, you can run
perl -e '@a = stat("."); print $a[11]'
which is just a convenient way to get the value of the 12th element of the
array returned by stat(2). On System V, this 12th element is a suggested or
optimal value of I/O size, meaning there could be a smaller I/O chunk size
(called fragment) which is the last, the 13th element ($a[12] above).
Steve Adams says that I/O buffer size is invariably 8k on UNIX. I tested the above Perl one liner on a number of UNIXes including FreeBSD. They're all 8192. On Linux, it's 4k. But many people don't think of Linux as UNIX.
In Windows, direct I/O (bypassing file system I/O) is used. The argument about I/O buffer size loses its weight and probably becomes irrelevant.
Note that the wide-spread notion that db_block_size should be chosen based on usage of the database is misleading. It's true only if you are using direct I/O, raw disks or some third-party file systems. On conventional file systems on UNIX, 8k is the best, regardless of whether you're building a DSS or OLTP database. This is just a paraphrase of Steve Adams' research. Search for his article at www.ixora.com.au/home.htm.
Yong Huang
yong321_at_yahoo.com
you wrote:
hi,
as my name suggests i am a novice.
i was reading ORACLE 8i DBA handbook by Kevin Loney and Marlene Theriault.
in the 5th chapter the author speaks of how I/O buffer cache affect the sizing.
My question is how do i findout what is my operating system's I/O buffer cache.
I work on Oracle 8.1.7 on
Windows NT 4.0 with SP5.
128 MB RAM
and
Redhat Linux 7.0
128MB RAM
thanks in advance
(note - i can not ask my system admin because there isn't one)
novice
 |
 |